As expected, the AI applications that have emerged in recent times will play a leading role at Codemotion Madrid 2026, which will take place at La Nave on April 20 and 21. In this article, we introduce the talk “Beyond SQL Generation: How to Teach Agents What Your Database Means.”
In this practical session, Kris Jenkins, Lead Developer Advocate at Snowflake, will explain why programming agents such as Claude struggle to retrieve relevant information from databases, what causes these limitations, and how to address them using semantic models and the OSI standard.
Context of the Talk
Modern AI models, particularly those behind Claude, have proven to be extremely effective at generating SQL code. However, there is a significant gap between producing syntactically correct queries and truly understanding the meaning of the underlying data.
Although these models excel at writing valid SQL, they often fall short where it matters most: delivering accurate and meaningful answers. This happens for a very simple reason—while their knowledge of SQL syntax is impressive, they lack awareness of the business context.
It is similar to hiring an expert in SQL syntax and expecting them to understand how your software works without ever explaining what each table represents or how the data is related. The result is perfectly written queries that return incorrect or irrelevant answers.
So, how can we enable an AI agent not only to write good SQL but also to truly understand the meaning of your database? The answer is straightforward: we must explicitly teach it the semantics of your data.
Semantic Models and the OSI Standard for AI Agents
What tools or techniques can provide an AI agent with a kind of guide that explains what each table means, how data entities relate to one another, what business rules apply, and what types of queries are meaningful? The answer lies in semantic models and the OSI (Open Semantic Interface) standard.
These technologies structure the meaning of data, the relationships between tables, and business rules, giving the agent the necessary context to generate useful queries. All the knowledge that an expert might acquire over six months can be provided to the agent in a structured file.
Semantic models act as an abstraction layer that defines business entities, their attributes, and relevant metrics independently of the underlying database structure.
The OSI standard, on the other hand, provides a machine-readable common format to express these semantic definitions, enabling AI agents to interpret them and generate more accurate and context-aware queries.
Today, it is already possible to create semantic models using specialized tools or frameworks that allow these layers to be defined declaratively.
Once implemented, this model connects natural language queries with the technical structure of the database, enabling the agent to understand not only how the data is organized but also what it represents in a business context.
The Talk
During the session, Kris Jenkins will demonstrate how to implement semantic models using the OSI standard, showing the difference between the responses produced by an agent without context and one that understands the semantics of the data.
Through live examples, attendees will see how to transform generic queries into truly useful results aligned with business needs.
He will also review the fundamentals of semantic models and the OSI standard, explain why the standard matters, and share techniques for quickly building effective semantic models. All of this with a single goal: to create a scalable, reliable, and effective database analyst from day one.
Kris will draw on his extensive experience at Snowflake, a leading cloud data platform company, where he has worked directly with teams facing these challenges on a daily basis.
His practical, results-oriented approach will ensure that attendees leave the session with concrete tools they can immediately apply in their own projects.
About the Speaker: Kris Jenkins, Lead Developer Advocate at Snowflake
Kris Jenkins is a lifelong programmer, Lead Developer Advocate at Snowflake, and host of the Developer Voices podcast. Throughout his career, he has served as CTO of a gold trading company, worked as a functional programming consultant specializing in Haskell, and regularly organized hackathons. He is passionate about solid software architecture, innovative programming languages, and, above all, building things.
